Individuals diagnosed with pathological gambling discounted hypothetical probabilistic monetary rewards significantly less steeply than matched control participants. A significant negative correlation between degree of probability discounting and scores on the SOGS was observed. However, unlike previous studies reported in the delay discounting literature (e.g., see review by Petry & Madden, in press
), pathological gamblers did not discount delayed monetary rewards significantly more steeply than matched controls.
The primary findings reported here are in accordance with those of Holt et al. (2003)
who found that college students with high SOGS scores (mean = 6.5) discounted hypothetical probabilistic monetary rewards less steeply than their peers with low SOGS scores (mean = 0.3). Our participants diagnosed with pathological gambling had much higher SOGS scores (range 7–20) than the student sample recruited by Holt et al., but the difference in degree of probability discounting separating gamblers from controls was qualitatively comparable. Quantitative comparisons cannot be made across studies because Holt et al. used much larger probabilistic reward amounts ($1,000-$50,000) than in the present study ($25-$85) and they did not report h
-values. Unlike the Holt et al. study, we observed no differences in degree of probability discounting across probabilistic reward amounts. This may be due to the substantially smaller differences in reward amounts in the present study (at most a 3.4-fold difference) when compared to the Holt et al. study (a 50-fold difference). Comparing our findings with those of Shead et al. (2008
, who also studied probability discounting in college-student gamblers) is more difficult because they did not include a sample of non-gambling controls. That they found no correlation between degree of probability discounting and scores on the Canadian Problem Gambling Index may owe to the high proportion of subjects classified as either low- or moderate-risk gamblers (81.4%) with smaller proportions of students classified as non-problem gamblers (8.5%) and problem gamblers (10.2%).
A second finding from the present study which is comparable to that reported by Holt et al. (2003)
is that although the group-averaged degree of delay discounting in the present study was higher for the gambling than the control group, this difference was not statistically significant (although when education and ethnicity were included in the ANOVA as covariates, the difference approached significance, p
= .058). The lack of a statistically significant effect of gambling status on degree of delay discounting is unusual in the delay discounting literature (Dixon, Marley, & Jacobs, 2003
; MacKillop et al., 2006
; Petry, 2001
; Petry & Casarella, 1999
). One difference between the present and past studies is that the sample size used here (38 participants combined) was small when compared to previous studies of delay discounting in pathological gamblers and matched controls (mean total participants = 64.5, range 47–86). This sample size may have also been too small for an assessment of group differences using the questionnaire developed by Kirby and Maraković (1995)
. For example, Kirby and Petry (2004)
found no significant group difference when the Kirby and Maraković questionnaire was completed by 33 alcohol-dependent and 44 control participants, although a number of studies, usually with larger samples, have found that alcohol-dependent individuals more steeply discount delayed rewards than controls (see review by Yi, Mitchell, & Bickel, in press
A final similarity between the present findings and those reported by Holt et al. (2003)
is that delay and probability discounting were not significantly correlated. This outcome is not unusual in the human delay/probability discounting literature. For example, Olson, Hooper, Collins, and Luciana (2007)
found no correlation between these types of discounting in 62 adolescents (8% of the variance in one type of discounting was accounted for by variance in the other). Myerson, Green, Hanson, Holt, and Estle (2003)
reported significant correlations between delay and probability discounting in two samples of 171 or 68 college students (the latter were data previously reported by Green, Myerson, & Ostaszewski, 1999
), but no significant correlation with a separate sample of 101 students. In all of the cases summarized by Myerson et al., the correlations were modest at best (Pearson’s r
ranged from .032 to .373).
Myerson et al. (2003)
note that when the correlation between delay and probability discounting is significant, the direction of that correlation is always positive. This would suggest a tendency for individuals that steeply discount delayed rewards to also heavily discount probabilistic rewards. As noted by Myerson et al., this positive correlation is counter-intuitive if one conceptualizes impulsivity as an inability to tolerate delays (steep delay discounting) and a propensity to take risks (shallow probability discounting). The extant delay discounting literature suggests that risk-taking pathological gamblers more steeply discount delayed rewards than controls, while the Holt et al. (2003)
study and the present findings suggest gamblers show more shallow probability discounting than controls. This suggests a negative correlation between delay and probability discounting in pathological gamblers might be observed. When we separately assessed the correlation between delay and probability discounting in our gamblers and controls, we found neither correlation was significant (not surprising given N = 19), however, the direction of these nonsignificant relations were opposite in pathological gamblers (r
= −.06) and controls (r
= .17). Future research with larger samples of pathological gamblers should investigate the possibility of a negative relation between probability and delay discounting in this population.
Two weaknesses of the present study which have not already been discussed deserve comment. First, the probability discounting questionnaire employed has not been used in prior studies of discounting. The questionnaire used was based on the delay discounting questionnaire developed by Kirby and Maraković (1995)
and intended to evaluate probability discounting along a similar range of reward magnitudes. Nonetheless, the probability discounting questionnaire had three more items than its delay discounting counterpart and was presented in blocks of choices with increasing win probabilities; whereas the delay discounting questionnaire was presented in a single block with a mixed sequence of delays. While these differences are potentially important, our participants’ choices on the probability discounting questionnaire were internally consistent in almost all cases.
A second weakness is that control participants were compensated $50 for transportation expenses while the pathological gamblers were not. While the gamblers received psychosocial treatment as a different form of compensation, the possibility remains that seeding the controls with $50 may have made them more risk prone. If this is so, then providing compensation may have decreased probability discounting, making them more like the pathological gamblers. Thus, the difference between these populations may have been bigger if this component of the procedure would have been held constant across groups.
Implications & Future Directions
Differences in the degree to which probabilistic rewards are discounted means either that the opportunity to gamble is subjectively worth more to pathological gamblers or that pathological gamblers display greater risk tolerance than controls. For example, given the rates of probability discounting obtained in this study, controls would be unwilling to forgo more than $20 to flip a coin on the chance of winning $80. However, our sample of pathological gamblers would be willing to forgo almost twice as much ($38, on average) on the same probabilistic outcome. Unanswered by this study is if pathological gamblers’ propensity for risky choice is confined to scenarios like those used here, or if it generalizes to other contexts. For example, are pathological gamblers more likely to forgo assured returns on an investment, preferring to take risks to obtain larger positive consequences such as stock market gains or improved health (e.g., through a surgical procedure that involves some risk)? Just as interesting is if more shallow probability discounting would be observed in pathological gamblers when the risky outcome involves probabilistic aversive events. Shead et al. (2008)
reported a significant negative correlation between probability discounting of gains and losses in their sample of college gamblers. Thus, those who placed a higher value on a probabilistic win (shallow probability discounting) tended to steeply discount the negative value of probabilistic losses (taking a “nothing bad will happen to me” stance). Shallow probability discounting of gains suggests gambling for gains is a valuable alternative. Steep discounting of probabilistic losses means that the individual is willing to forgo very little (a certain payment) to avoid rolling the dice on a probabilistic loss. If this negative correlation is a general tendency across conditions and outcomes, pathological gamblers would, on average, be expected to more steeply discount the negative value of contracting a sexually transmitted disease by engaging in risky sexual practices. Petry (2000)
noted that problem-gambling substance abusers engaged in significantly greater risky behaviors that spread HIV and contagious diseases than their non-problem gambling counterparts. Further study of probabilistic discounting and its impact on behavioral decision making in pathological gamblers is needed.
A second interesting question raised but not answered by the present research is if shallow probability discounting precedes and predicts pathological gambling. Some evidence obtained from nonhuman experiments suggests that degree of delay
discounting is predictive of acquisition of cocaine self-administration and reinstatement of nicotine self-administration in rodents (see review by Carroll, Anker, Mach, Newman, & Perry, in press
). Conducting a similar experiment by assessing probability discounting and then examining subsequent development of pathological gambling is hampered by at least two factors. First, conducting the required longitudinal research with humans may be impractical given the low incidence rate of pathological gambling (Petry, Stinson, & Grant 2005
; Kessler et al., 2008
). Second, although animal research has the advantage of controlling extraneous factors that may affect probability discounting and gambling, at present animal preparations have yet to adequately capture the functional characteristics of human gambling. For example, human gamblers wager, win, and lose, token reinforcers whereas animal studies have thus far been unsuccessful in establishing a surplus of token reinforcers which an animal might wager unless exchanging tokens for food is restricted by requiring the subject complete a large work requirement before tokens may be exchanged (Yankelevitz, Bullock, & Hackenberg, 2008
). This is important because gambling losses which mirror gambling wins cannot be arranged with food reinforcers which, once consumed, cannot be lost (see Madden, Ewan, & Lagorio, 2007
In sum, the present data suggest that pathological gamblers discount probabilistic rewards significantly less steeply than controls. While a trend toward greater delay discounting was noted in the pathological gamblers, the between group differences were much more robust for probabilistic discounting. These results point to potentially unique aspects of decision making in pathological gamblers that may help explain the onset and/or progression of the disorder. Further research is needed to replicate and extend these findings, and to determine interventions that may assist in preventing individuals with shallow probabilistic discounting functions from developing gambling problems or in abating gambling problems among those who have developed the disorder.